A neonatal intensive care unit service platform

Abstract

The application discloses a new-born intensive care unit service platform and relates to the technical field of medical informatization. The application performs semantic analysis on the consultation question about the new-born proposed by the family member, judges whether the corresponding reply can be automatically generated based on a preset medical knowledge base, and if not, transfers the question to an artificial reply end for artificial answering by medical staff. The state information of the medical staff at the artificial reply end is acquired to calculate a reply priority value. The consultation question is sent to the medical staff corresponding to the maximum reply priority value for answering the consultation question. In this way, when the system cannot answer the question of the family member and needs to be transferred to the corresponding medical staff for artificial reply, the suitable medical staff can be reasonably selected according to the actual situation to reply, information transmission is ensured to be timely, the reply content is accurate, the anxiety of the family member is reduced, the trust relationship between the doctor and the patient is increased, and the adverse effect on the treatment of the new-born is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical informatization, and particularly relates to a neonatal intensive care unit service platform. BACKGROUND

[0002] The neonatal intensive care unit (NICU) is an important medical unit for treating critically ill newborns, but due to its special nature of closed management, there are many obstacles in the information communication between family members and medical staff; in order to be able to answer the doubts of family members, and at the same time help family members to timely discover the problems of newborns, the service platform sets an intelligent question and answer system, specifically through the intelligent question and answer system to realize the interaction between family members and the platform. The system uses natural language processing technology to analyze and understand the questions raised by the family members, combines with the medical knowledge base and the pre-set rules, and automatically gives accurate and timely replies. For the questions that the system cannot answer, it is transferred to the corresponding medical staff for manual reply to ensure that the problems are properly solved;

[0003] However, in the existing intelligent question and answer system, when the system cannot answer the questions of the family members and needs to be transferred to the corresponding medical staff for manual reply, if the suitable medical staff cannot be reasonably selected according to the actual situation to reply, it may lead to the problems of not timely information transmission or inaccurate reply content, thereby intensifying the anxiety of the family members, affecting the trust relationship between doctors and patients, and even possibly causing adverse effects on the treatment of newborns due to the delay of important information transmission. SUMMARY

[0004] The purpose of the present application is to solve the above-mentioned problems, and to provide a neonatal intensive care unit service platform.

[0005] In the first aspect of the present application, a neonatal intensive care unit service platform is first proposed, which comprises an intelligent question and answer system, and the intelligent question and answer system comprises:

[0006] An analysis and judgment module analyzes and judges the semantic analysis of the consultation questions raised by the family members about the newborns, and judges whether the corresponding reply can be automatically generated based on the pre-set medical knowledge base;

[0007] An artificial transfer module: if the corresponding reply cannot be automatically generated, the question is transferred to the artificial reply end for artificial answering by the medical staff;

[0008] A reply selection priority module obtains the state information of the medical staff of the artificial reply end, and calculates a reply priority value according to the state information;

[0009] A question reply module sends the consultation question to the medical staff corresponding to the maximum reply priority value as the target medical staff, and the target medical staff replies to the consultation question.

[0010] Optionally, the step of performing semantic analysis on the consultation question about the newborn raised by the family member and determining whether an automatic reply can be generated based on the preset medical knowledge base is:

[0011] The content of the consultation question about the newborn raised by the family member is subjected to word segmentation processing, and semantic information reflecting the core meaning of the question is extracted in combination with part-of-speech tagging and named entity recognition technology.

[0012] The extracted question semantics are subjected to vectorization processing based on a pre-trained BERT language model to generate corresponding semantic vector representations.

[0013] The semantic vector is compared with the standard question semantic vector that has been subjected to vectorization processing in the preset medical knowledge base in terms of similarity, and the cosine similarity algorithm is used for similarity calculation.

[0014] If the similarity between the semantic vector and the semantic vector of a certain standard question is higher than a set similarity threshold, the standard question is determined to be the most matched question, and the corresponding standard answer is selected as the automatic reply content of the family consultation question.

[0015] If the similarity between the semantic vector and the semantic vector of any standard question in the knowledge base does not reach the set threshold, it is determined that the current question is a question that cannot be automatically replied by the knowledge base, and the question is transferred to the artificial transfer module for processing.

[0016] Optionally, the step of obtaining the state information of the medical staff at the artificial reply end and calculating the reply priority value according to the state information is:

[0017] The state information of the medical staff includes a reply timeliness coefficient, a reply time stability coefficient, a fatigue coefficient, and a reply experience coefficient, and the reply priority value is calculated according to the reply timeliness coefficient, the reply time stability coefficient, the fatigue coefficient, and the reply experience coefficient.

[0018] Optionally, the step of calculating the reply timeliness coefficient is:

[0019] The number of times that each medical staff at the artificial reply end has historically completely replied to the consultation question about the newborn raised by the family member is obtained, and the interval between the time when the family member raises the question and the time when the user replies to the question in each reply process is obtained, and the mean of the interval is calculated as the reply feedback time in the corresponding reply process.

[0020] The reply feedback time in each reply process is marked as , , where n represents the number of reply processes, =1, 2, 3, 4, …, , is a positive integer.

[0021] According to the reply feedback time in each reply process and the preset slowest reply feedback time Calculate the reply timeliness coefficient, and the formula is: , wherein, is the reply timeliness coefficient.

[0022] Optionally, the calculation step of the reply time stability coefficient is:

[0023] From the number of times that each medical staff at the artificial reply end historically completely replies to the consultation question about the newborn raised by the family members, and the interval between adjacent two questions in each reply process is obtained, and the interval between each adjacent two questions is marked as , represents the order number of the interval between each adjacent two questions, =1, 2, 3, 4, …, , and is a positive integer;

[0024] Calculate the standard deviation of the interval between adjacent two questions in each reply process, and mark the standard deviation as , the standard deviation is calculated according to the formula: , wherein, is the average value of the interval between adjacent two questions in each reply process, and the expression for obtaining is: ;

[0025] According to the standard deviation in each reply process, calculate the reply time stability coefficient, and the formula is: , wherein, is the reply time stability coefficient, represents the number of the reply process, =1, 2, 3, 4, …, , is a positive integer.

[0026] Optionally, the calculation step of the fatigue coefficient is:

[0027] Obtain the total time of each medical staff participating in replying to the consultation question about the newborn raised by the family members in the recent preset time period from the artificial reply end as the reply duration, and divide the reply duration by the preset time period to obtain the reply time proportion;

[0028] Obtain the total number of the unanswered reply questions of each medical staff, and normalize the total number of the unanswered reply questions to map to the numerical interval of 0-1;

[0029] Obtain the time from the last rest of each medical staff, and normalize the time from the last rest to map to the value interval of 0-1;

[0030] According to the total number of unfinished reply questions after normalization processing, the time from the last rest and the reply time ratio, the fatigue coefficient is calculated, and the formula is: In the formula, is the fatigue coefficient, , and respectively, the total number of unfinished reply questions after normalization processing, the time from the last rest and the reply time ratio, respectively, , and are preset proportion coefficients, and , and are greater than 0.

[0031] Optionally, the calculation step of the reply experience coefficient is:

[0032] Obtain the number of times of historical complete reply of each medical staff to the consultation questions about newborns raised by family members from the artificial reply end, and obtain the time interval from the time of starting to reply questions to the current time, denoted as the time interval of replying;

[0033] Divide the number of times of complete reply to the consultation questions about newborns raised by family members by the time interval of replying, denoted as the reply density value;

[0034] Obtain the reply satisfaction score of each medical staff after each reply is completed, and calculate the mean value of the reply satisfaction score as the reply satisfaction degree;

[0035] Normalize the reply density value and the reply satisfaction degree, and calculate the reply experience coefficient according to the normalized reply density value and the reply satisfaction degree, and the formula is: In the formula, is the reply experience coefficient, and respectively, the normalized reply density value and the reply satisfaction degree, is a preset weight factor, and the value range is in the interval of 0-1.

[0036] Optionally, the step of calculating the reply priority value according to the reply timely coefficient, the reply time smooth coefficient, the fatigue coefficient and the reply experience coefficient is:

[0037] The reply timely coefficient, the reply time stable coefficient, the fatigue coefficient and the reply experience coefficient are normalized to calculate the reply priority value, and the steps of calculating the reply priority value according to the normalized reply timely coefficient, the reply time stable coefficient, the fatigue coefficient and the reply experience coefficient are as follows:

[0038] In the formula, is the reply priority value, are the normalized reply timely coefficient, the reply time stable coefficient, the fatigue coefficient and the reply experience coefficient respectively, are respectively a preset proportion coefficient of the reply timely coefficient, the reply time stable coefficient, the fatigue coefficient and the reply experience coefficient, and are all greater than 0.

[0039] The present application has the following advantages:

[0040] The present application provides a new-born intensive care unit service platform, which performs semantic analysis on the consultation question about the new-born proposed by the family member, and determines whether the corresponding reply can be automatically generated based on a preset medical knowledge base; if the corresponding reply cannot be automatically generated, the question is transferred to an artificial reply end for artificial answering by medical staff; the state information of the medical staff of the artificial reply end is acquired, and the reply priority value is calculated according to the state information; the consultation question is sent to the medical staff corresponding to the maximum reply priority value as the target medical staff, and the target medical staff replies to the consultation question. In this way, when the system cannot answer the question of the family member and needs to be transferred to the corresponding medical staff for artificial reply, the suitable medical staff can be reasonably selected for reply according to the actual situation, the information transmission is ensured to be timely, the reply content is ensured to be accurate, the anxiety of the family member is reduced, the trust relationship between the medical staff and the patient is increased, and the adverse effect on the treatment of the new-born is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0041] The present application will be further described below in conjunction with the drawings.

[0042] Figure 1 It is a framework diagram of a new-born intensive care unit service platform. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0044] The present application provides a new-born intensive care unit service platform. Referring to Figure 1 , Figure 1A framework diagram of a neonatal intensive care unit service platform provided by an embodiment of the present application. It includes an intelligent question and answer system, which includes:

[0045] An analysis and judgment module performs semantic analysis on the consultation question about the neonate raised by the family member, and determines whether an automatic reply can be generated based on a preset medical knowledge base;

[0046] An artificial switching module: if an automatic reply cannot be generated, the question is switched to an artificial reply end, and medical staff perform artificial answering;

[0047] A reply selection priority module acquires state information of the medical staff of the artificial reply end, and calculates a reply priority value according to the state information;

[0048] A question reply module sends the consultation question to the medical staff corresponding to the maximum reply priority value as target medical staff, and the target medical staff replies to the consultation question.

[0049] Based on the neonatal intensive care unit service platform provided by the embodiment of the present application, when the system cannot answer the question of the family member and needs to be switched to the corresponding medical staff for artificial reply, the suitable medical staff can be reasonably selected for reply according to the actual situation, the information transmission is ensured to be timely, the reply content is accurate, the anxiety of the family member is reduced, the trust relationship between the doctor and the patient is increased, and the adverse effects on the treatment of the neonate are reduced.

[0050] In one embodiment, the step of performing semantic analysis on the consultation question about the neonate raised by the family member and determining whether an automatic reply can be generated based on a preset medical knowledge base is:

[0051] The content of the consultation question about the neonate raised by the family member is processed by word segmentation, and combined with part-of-speech tagging and named entity recognition technology, semantic information reflecting the core meaning of the question is extracted;

[0052] The extracted question semantics are vectorized based on a pre-trained BERT language model to generate corresponding semantic vector representations;

[0053] The semantic vector and the standard question semantic vector that has been vectorized in the preset medical knowledge base are compared in similarity, and the cosine similarity algorithm is used for similarity calculation;

[0054] If the similarity between the semantic vector and the semantic vector of a certain standard question is higher than a set similarity threshold, the standard question is determined to be the most matched question, and the corresponding standard answer is selected as the automatic reply content of the family consultation question;

[0055] If the similarity of the semantic vector to any standard question in the knowledge base does not reach the set threshold, it is determined that the current question is a question that cannot be automatically replied by the knowledge base, and the question is transferred to the artificial transfer module for processing.

[0056] It should be noted that in order to realize accurate understanding and intelligent response to the new-born baby consultation question raised by the family member, the system first performs natural language processing on the input question, specifically including word segmentation processing, part-of-speech tagging, and application of named entity recognition (NER) technology. Word segmentation processing splits long sentences into basic semantic units; part-of-speech tagging identifies the grammatical roles of each word, such as nouns, verbs, adjectives, etc.; named entity recognition further extracts medical entity information, such as disease names (such as “new-born baby jaundice”), symptoms (such as “rapid breathing”), drug names, etc. Next, the system performs deep semantic modeling on the extracted semantic information based on the BERT language model pre-trained on a large amount of medical text to generate a semantic vector representation of the question. Subsequently, the system compares the semantic vector with the vector of the standard question pre-constructed in the medical knowledge base for semantic similarity, and uses the cosine similarity algorithm to calculate the similarity score between the two. For example, for the question “Is it that the baby always spits milk after feeding? Is the stomach not good?” raised by the family member, the system identifies the key words “feeding” “spitting milk” and “stomach not good” after word segmentation, and identifies “spitting milk” as a symptom and “stomach” as an organ entity, and generates its semantic vector after BERT encoding. The system compares the vector with the semantic vectors of standard questions such as “What are the reasons for frequent spitting milk in new-born babies?” and “Is it normal for a baby to spit milk?” in the knowledge base. If the similarity between it and “What are the reasons for frequent spitting milk in new-born babies?” is 0.89 (higher than the set similarity threshold of 0.8), it is determined as the most matched question, and the system will return the standard answer corresponding to the standard question, such as “New-born baby spitting milk is common, which may be related to feeding posture, excessive food intake or immature digestive system development, and it is recommended to adjust the feeding method and observe the situation.” as the automatic reply content.

[0057] If the comparison result shows that the similarity of the question to all standard questions does not reach the threshold (such as being lower than 0.8), it is considered that the question is complex or not clearly expressed, and the system cannot determine a clear reply content, thereby triggering the artificial transfer mechanism to transfer the question to medical staff for further processing. This processing flow not only guarantees the accuracy of the automatic reply of the system, but also provides a safety mechanism for artificial intervention for complex questions.

[0058] In one implementation, through the above process, the intelligent question and answer system not only guarantees the automation efficiency, but also effectively avoids the risk of incorrect or inaccurate reply, thereby improving user trust and system reliability.

[0059] In one embodiment, the steps of obtaining the status information of the medical staff at the human response terminal and calculating the response priority value based on the status information are as follows:

[0060] The status information of medical staff includes response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient. The response priority value is calculated based on the response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient.

[0061] In one embodiment, the steps for responding to the timeliness coefficient are as follows:

[0062] The number of times that medical staff at the human response end have completely answered family members’ inquiries about newborns is recorded. During each response, the interval between the time the parent asks the question and the time the user answers the question is obtained, and the average interval is calculated as the response feedback time in the corresponding response process.

[0063] Mark the response time for each response process as . , The number indicating the number of response steps. =1, 2, 3, 4, ... , It is a positive integer;

[0064] Based on the response time and the preset slowest response time during each response process. The formula for calculating the timeliness coefficient of response is as follows: In the formula, The response time factor.

[0065] It should be noted that the data used in the above-mentioned calculation of the response timeliness coefficient mainly comes from the interaction logs of the manual response system. Each time a family member initiates a consultation, the system automatically records the timestamp of the question submission; when the corresponding medical staff responds, the system also records the response timestamp. By analyzing these timestamp pairs, the system can accurately calculate the response time for each reply. All historical response records are stored in the database. The system retrieves the historical response logs of each medical staff member, extracts the time information, categorizes them by question number, and calculates the average time interval for each response process. Simultaneously, a preset minimum response time can be set by the hospital or platform according to service standards, serving as a benchmark for calculating the response timeliness coefficient. Through this structured time data, the system can achieve a quantitative assessment of response efficiency.

[0066] It should be noted that the reply timeliness coefficient refers to the time efficiency index of the medical staff in responding to the new-born baby consultation questions raised by the family members in the historical manual reply process, which is specifically manifested as the ratio or mapping value between the average reply feedback time and the slowest acceptable reply time set by the platform. Its essence is to evaluate whether the medical staff can respond to the family members' questions in a short time, reflecting the speed of handling the problem. The larger the reply timeliness coefficient is, the shorter the average response time of the medical staff in the past reply is, and the stronger the time sensitivity and service efficiency are, so it is more reliable. In the intelligent question and answer system, when it is not possible to automatically generate a standard reply, the problem needs to be quickly and effectively transferred to the manual reply end. At this time, if a medical staff who replies quickly and is used to quick response is selected, not only the anxiety of the family members caused by waiting can be effectively alleviated, but also the overall service satisfaction of the platform can be improved. For example, assuming that there are two medical staff A and B in the system, A replies on average within 1 minute, B replies on average within 5 minutes, and the platform sets the slowest acceptable time as 10 minutes, then the reply timeliness coefficient of A is 0.9 and that of B is 0.5. When transferring, the system will preferentially assign the problem to A because its reply priority value is higher and it can respond to the user faster to ensure efficient and smooth information communication. This mechanism can significantly reduce the risk of information delay, which is particularly crucial for the NICU, which is a high-risk and information time-sensitive environment.

[0067] In one implementation, calculating the reply timeliness coefficient has important significance and practical benefits for determining which medical staff to transfer the problem that cannot automatically generate a corresponding reply to. First, it makes the problem transfer process more accurate and efficient, and by preferentially selecting medical staff who have historically responded quickly, it can significantly shorten the time for family members to wait for a reply, alleviate their nervous and anxious emotions, and improve the communication experience between doctors and patients. Second, in the NICU environment, family members often pay close attention to every change in the new-born baby. Once a problem occurs, if the system cannot immediately transfer it to a medical staff with high reply timeliness, it may cause a delay in information communication, which in turn affects the subsequent diagnosis and treatment rhythm. After introducing the reply timeliness coefficient, the platform can make intelligent decisions based on the quantitative efficiency index, so that the problem is always preferentially transferred to medical staff who handle responses quickly and have a strong sense of time, improving the overall reply quality and service efficiency. For example, during the night or when the number of staff is relatively tight, the system can still intelligently select the most responsive personnel based on the reply timeliness coefficient, avoiding missing critical reply windows due to human judgment errors, and effectively ensuring the continuity of new-born baby care and the information right of family members.

[0068] In one embodiment, the calculation step of the reply time stability coefficient is:

[0069] the number of times each medical staff at the artificial answering end completely answers the consultation questions raised by the family members about the newborns, and the interval between adjacent two questions in each answering process is obtained, and the interval between each adjacent two questions is marked as , the order number of the interval between each adjacent two questions, = 1, 2, 3, 4, …, , and is a positive integer;

[0070] the standard deviation of the interval between adjacent two questions in each answering process is calculated, and the standard deviation is marked as , the calculation formula of the standard deviation is as follows: wherein, is the average value of the interval between adjacent two questions in each answering process, and the expression for obtaining the average value is as follows: ;

[0071] the answering time stability coefficient is calculated according to the standard deviation of each answering process, and the calculation formula is as follows: , wherein, is the answering time stability coefficient, is the number of the answering process, = 1, 2, 3, 4, …, , is a positive integer.

[0072] It should be noted that the data involved in the above answering time stability coefficient calculation process can be automatically obtained through the answering interaction log recorded in the artificial answering system. Specifically, the system will record the time stamp of the interaction between each medical staff and the family members in each answering process, including the time points of each question and the corresponding answer. By analyzing the difference between these time points, the time interval between adjacent two questions in each answering process can be extracted, and a series of interval sequences can be generated. The system further calculates the average time interval of each answering process and the corresponding standard deviation based on these time interval data, and further derives the answering time stability coefficient of each medical staff. These data do not need to be manually entered, and the system automatically monitors and collects them, so that the historical answering times, answering time intervals, time distribution rules and other information can be obtained, which provides accurate data support for subsequent standard deviation calculation and stability evaluation, thereby improving the automation and intelligent level of answering service evaluation.

[0073] It should be noted that the reply time smoothness coefficient is an index for measuring the stability of the reply time interval of the medical staff when replying to the user's inquiry question, reflecting whether the time rhythm in multiple reply processes is uniform, and whether the reply has good continuity and concentration. The coefficient calculates the standard deviation of the time interval between adjacent two questions in each manual reply, and further evaluates the time fluctuation of the overall reply behavior based on these standard deviations. The larger the reply time smoothness coefficient is, the more stable the time rhythm of the medical staff in different reply processes is, and it is not easy to have long interruption or delay in reply, which represents that the medical staff has strong reply concentration ability and operation continuity, and can more efficiently and timely handle user questions, so the priority of manual access reply of the medical staff should be relatively higher. In the situation where accurate replies cannot be automatically generated, the problem is preferentially transferred to the medical staff with high time smoothness, which can improve the overall reply quality and response efficiency and enhance user satisfaction. For example, the reply time intervals of medical staff A are 8 seconds, 9 seconds, 10 seconds and 9 seconds, the standard deviation is small, and the smoothness coefficient is large; while the reply intervals of medical staff B are 5 seconds, 20 seconds, 7 seconds and 18 seconds, the fluctuation is large, and the smoothness coefficient is small. Therefore, when transferring manual reply, the problem should be assigned to A in priority.

[0074] In an implementation, calculating the reply time smoothness coefficient has significant practical value for determining which medical staff to transfer the problem that cannot be automatically replied to. By evaluating the time stability of the medical staff in historical replies, the system can identify those who can maintain a stable reply rhythm when facing multiple rounds of dialogue. Such personnel usually have good attention allocation ability, skilled reply skills and strong service consciousness, and can handle multiple complex or urgently manual intervention problems in a short time. Prioritizing the problems that the automatic reply system cannot cover and transferring them to the medical staff with a higher reply time smoothness coefficient can help ensure the continuity and timeliness of the manual reply process, reduce waiting time, and improve user experience and satisfaction. For example, when a problem involves a sudden health condition of a newborn baby and the family members have a large emotional fluctuation, fast and professional manual feedback is needed, and the problem should be assigned to the medical staff with high time smoothness in priority, which can effectively avoid interruption or delay in the reply process, thereby ensuring the reply quality and communication efficiency.

[0075] In one embodiment, the calculation step of the fatigue coefficient is:

[0076] The total time of each medical staff participating in replying to the inquiry question about the newborn baby raised by the family member in the recent preset time period is obtained as the reply duration from the manual reply end, and the reply duration is divided by the preset time period to obtain the reply time proportion;

[0077] obtain the total number of unanswered questions of each medical staff, and normalize the total number of unanswered questions to map to the numerical interval of 0-1;

[0078] obtain the time from the last rest of each medical staff, and normalize the time from the last rest to map to the numerical interval of 0-1;

[0079] According to the total number of unanswered questions, the time from the last rest, and the reply time proportion after normalization, the fatigue coefficient is calculated, and the formula is: In the formula, is the fatigue coefficient, , and are the total number of unanswered questions, the time from the last rest, and the reply time proportion after normalization, respectively, , and are preset proportion coefficients, and , and are all greater than 0.

[0080] It should be noted that in the calculation process of the fatigue coefficient, the data involved is mainly recorded and retrieved in real time by the background system of the artificial reply terminal platform. Specifically, when the medical staff replies to the family member's consultation question, the system automatically records the start and end time of each reply, accumulates and calculates the total reply time in the recent preset time period (such as the last 6 hours or 8 hours) to determine the "reply duration"; at the same time, the platform continuously tracks the number of questions currently received but not completed by each medical staff as the "total number of unanswered questions", and normalizes it through the standard maximum range; in addition, the system also records the time node of the last rest state of each medical staff, and calculates the difference value with the current time to obtain the "time from the last rest", which is also normalized to the interval of 0-1. These data are automatically collected and sorted by the system log, scheduling management system, task distribution record and other modules of the platform, ensuring the dynamic and accuracy of the fatigue coefficient, and providing comprehensive and real-time support basis for the system to judge the current fatigue degree of the medical staff.

[0081] It is to be noted that the fatigue coefficient refers to the degree of fatigue exhibited by the medical staff due to factors such as excessive work duration, excessive number of unanswered questions, and excessive time since the last rest. The larger the value of the fatigue coefficient, the higher the workload and fatigue level of the medical staff. Conversely, the smaller the fatigue coefficient, the lighter the workload and the more energetic the medical staff. In this case, the medical staff with a smaller fatigue coefficient have relatively higher ability and efficiency in handling new problems and answering family questions, so the system will prioritize transferring problems that cannot be automatically replied to these less-fatigued medical staff for manual answering. For example, assume that medical staff A has handled 30 consultation questions in the past 4 hours and has not had enough rest, while medical staff B has handled only 10 questions in the same time period and has recently had better rest. If medical staff A has a higher fatigue coefficient, it means that they may be less efficient, slow in thinking, and even make mistakes when answering questions, while medical staff B has a lower fatigue coefficient and is in good working condition, able to efficiently and accurately answer questions raised by family members. In this case, the system is more inclined to transfer problems that cannot be automatically replied to medical staff B rather than A, which helps to ensure that family questions are answered in a timely and professional manner. In this way, the risk of declining response quality due to fatigue of medical staff is reduced, and the response efficiency of the system is improved.

[0082] In one implementation, calculating the fatigue coefficient is beneficial for determining which medical staff to transfer the problem that cannot be automatically replied to, as it helps to ensure the quality and efficiency of the response to the problem. The fatigue coefficient reflects the current mental state and working capacity of the medical staff by quantifying their workload and rest. If the system can prioritize transferring problems to medical staff with a lower fatigue coefficient, it can to some extent avoid the decline in judgment and inaccurate responses due to excessive work. This not only improves the accuracy and timeliness of the response, but also reduces the psychological pressure of the medical staff due to excessive fatigue, avoiding their work omissions caused by fatigue. At the same time, reasonable allocation of workload helps to improve the job satisfaction of medical staff and the overall work efficiency of the team, ultimately improving the medical experience of family members and strengthening the trust relationship between doctors and patients. Therefore, the calculation of the fatigue coefficient provides an effective tool for the workload management of medical staff, thereby helping the entire system to operate more intelligently and efficiently.

[0083] In one embodiment, the step of calculating the reply experience coefficient is:

[0084] The number of times each medical staff from the manual reply end has completely answered the consultation questions raised by family members about the newborn, and the time interval from the start of answering questions to the current time, denoted as the time interval for answering questions, is obtained.

[0085] The number of times that the family's consultation questions about the newborn are completely answered is divided by the time interval in which the answering is done, denoted as the reply density value;

[0086] The reply satisfaction score after each reply of each medical staff is obtained, and the mean value of the reply satisfaction score is calculated as the reply satisfaction degree;

[0087] The reply density value and the reply satisfaction degree are normalized, and the reply experience coefficient is calculated according to the normalized reply density value and the normalized reply satisfaction degree, and the formula is: , wherein, is the reply experience coefficient, and are the normalized reply density value and the normalized reply satisfaction degree, respectively, is a preset weight factor, and the value range is in the interval of 0-1. Generally, the value is 0.6;

[0088] It should be noted that the data acquisition method involved in the above reply experience coefficient calculation process mainly includes three aspects. First, regarding the "number of times that the family's consultation questions about the newborn are completely answered" and "time interval in which the answering is done", these data can be extracted from the system log of the artificial reply end. The system should record the time stamp of each medical staff participating in the reply and the number of family questions, and then calculate the reply density value. Second, regarding the "reply satisfaction score", these data are usually obtained from the feedback of the family to the reply, and can be collected through the platform after each reply, or obtained through questionnaires, satisfaction surveys and other ways. Finally, regarding the normalization of the "reply density value" and the "reply satisfaction degree", the standardization method can be used, such as subtracting the minimum value from each data point and dividing by the difference between the maximum value and the minimum value, so as to map the data to the interval of 0-1. Through the integration and calculation of these data, the reply experience coefficient of each medical staff can be obtained, which provides effective support for the intelligent transfer system.

[0089] It should be noted that the reply experience coefficient is a comprehensive index for measuring the experience and ability of medical staff in handling family members' consultation problems about newborns. The coefficient integrates two key factors: reply density and reply satisfaction. Reply density reflects the frequency of medical staff handling problems within a certain time, while reply satisfaction reflects the evaluation of family members on the quality of medical staff's reply. By normalizing these two indicators and calculating the weighted value, the reply experience coefficient is obtained. The larger the reply experience coefficient, the more consultation problems medical staff can effectively and high-quality handle in a short time, indicating higher experience and coping ability. When the reply experience coefficient is high, it means that the medical staff has strong ability and high satisfaction in handling consultation problems, which enables the medical staff to answer the problems raised by family members more quickly and accurately. Therefore, when the system decides whether to transfer the problems that cannot be automatically replied to human, it will give priority to medical staff with a high reply experience coefficient to ensure that the answer after transfer is more timely and of higher quality. For example, if a medical staff has handled a large number of newborn-related consultation problems in the past period of time and has a high reply satisfaction, the system will consider that the medical staff has rich experience and can efficiently solve the problems of family members, thus preferentially transferring complex or automatically unreplied problems to him / her.

[0090] In one implementation, calculating the reply experience coefficient has the advantage of improving the efficiency and quality of transfer for the following reasons. By considering the reply density and reply satisfaction of medical staff, the system can accurately assess the experience and ability of each medical staff. For problems that cannot be automatically replied, the system can preferentially transfer these problems to medical staff with rich experience and high reply quality, thereby improving the efficiency of solving family problems and reducing the risk of declining reply quality due to transfer to inexperienced personnel. For example, if a medical staff has a high reply satisfaction and a high reply density in history, indicating that he / she can quickly and accurately answer problems, the system can preferentially select him / her for reply, which can greatly reduce the waiting time of family members and improve their satisfaction. This transfer strategy based on experience coefficient not only optimizes resource allocation, but also enhances the quality and efficiency of overall service.

[0091] In one embodiment, the step of calculating the reply priority value according to the reply timeliness coefficient, the reply time stability coefficient, the fatigue coefficient, and the reply experience coefficient is:

[0092] The step of calculating the reply priority value according to the normalized reply timeliness coefficient, the normalized reply time stability coefficient, the normalized fatigue coefficient, and the normalized reply experience coefficient is:

[0093] In the formula, is the answer priority value, respectively, are the normalized answer and time coefficient, answer time stability coefficient, fatigue coefficient and answer experience coefficient, respectively, is the preset proportion coefficient of are all greater than 0.

[0094] It should be noted that the consultation question is sent to the medical staff corresponding to the maximum answer priority value as the target medical staff, and the target medical staff answers the consultation question. After the consultation question is sent to the medical staff corresponding to the maximum answer priority value, the target medical staff will answer the question in detail according to the consultation content provided by the system and his professional knowledge. The medical staff provides accurate solutions or suggestions according to his experience, professional background and understanding of the question. At the same time, the system will monitor the response time and answer quality in the answering process in real time to ensure that the medical staff completes the answer within the specified time and provides further guidance or supplementary information as needed. In this way, the consultation question can be quickly and accurately solved, improving the efficiency of medical services and enhancing the satisfaction of patients.

[0095] The above describes one embodiment of the present application in detail, but the content is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the scope of the present application.

Claims

1. A neonatal intensive care unit service platform, characterized in that, The service platform includes an intelligent question-and-answer system, which includes: The analysis and judgment module performs semantic parsing on the questions raised by family members about newborns, and determines whether a corresponding response can be automatically generated based on a preset medical knowledge base; Manual transfer module: If a corresponding response cannot be generated automatically, the question will be transferred to the manual response end for medical staff to answer it manually. The response selection priority module obtains the status information of medical staff at the human response end and calculates the response priority value based on the status information; The question and answer module dispatches inquiries to the medical staff with the highest priority value for answering, who then answer the inquiries. The steps for obtaining the status information of medical staff at the human response end and calculating the response priority value based on the status information are as follows: The status information of the medical staff includes response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient. The response priority value is calculated based on the response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient. The steps for calculating the response time stationarity coefficient are as follows: The system retrieves the number of times medical staff at the human response end have completely answered family inquiries about newborns, and records the interval between two adjacent questions during each response. This interval is then marked as [missing information]. , This indicates the sequential numbering of the interval between any two adjacent questions. =1, 2, 3, 4, ... ,and It is a positive integer; Calculate the standard deviation of the interval between two adjacent questions during each response process, and label the standard deviation as... Standard deviation The calculation formula is: in, The expression for the average interval between two adjacent questions during each response process is: ; Based on the standard deviation during each response The formula for calculating the stationarity coefficient of response time is as follows: In the formula, To provide the time stationarity coefficient, The number indicating the number of response steps. =1, 2, 3, 4, ... , It is a positive integer.

2. The neonatal intensive care unit service platform according to claim 1, characterized in that, The steps for semantically parsing family members' inquiries about newborns and determining whether a corresponding response can be automatically generated based on a pre-set medical knowledge base are as follows: The questions raised by family members regarding newborns are processed by word segmentation, and semantic information reflecting the core meaning of the questions is extracted by combining part-of-speech tagging and named entity recognition technology. The extracted question semantics are vectorized based on the pre-trained BERT language model to generate the corresponding semantic vector representation. The semantic vector is compared with the standard question semantic vector in the preset medical knowledge base that has been vectorized. The similarity calculation method adopts the cosine similarity algorithm. If the similarity between the semantic vector and the semantic vector of a certain standard question is higher than the set similarity threshold, then the standard question is determined to be the best matching question, and its corresponding standard answer is selected as the automatic reply content for the family consultation question. If the similarity between the semantic vector and any standard question in the knowledge base does not reach the set threshold, the current question is determined to be a question that cannot be automatically answered by the knowledge base, and the question is transferred to the manual transfer module for processing.

3. The neonatal intensive care unit service platform according to claim 1, characterized in that, The steps for determining the timeliness of the response are as follows: The number of times that medical staff at the human response end have completely answered family members’ inquiries about newborns is recorded. During each response, the interval between the time the parent asks the question and the time the user answers the question is obtained, and the average interval is calculated as the response feedback time in the corresponding response process. Mark the response time for each response process as . , The number indicating the number of response steps. =1, 2, 3, 4, ... , It is a positive integer; Based on the response time and the preset slowest response time during each response process. The formula for calculating the timeliness coefficient of response is as follows: In the formula, The response time factor.

4. The neonatal intensive care unit service platform according to claim 1, characterized in that, The steps for calculating the fatigue coefficient are as follows: The total time spent by each medical staff member in answering family members' inquiries about newborns within the most recent preset time period is obtained from the manual response terminal. This is taken as the response duration, and the response duration is divided by the preset time period to obtain the response time percentage. Obtain the total number of unanswered questions for each medical staff member, and normalize the total number of unanswered questions to a value range of 0-1; Obtain the time since each medical staff member last rest, and normalize the time since the last rest, mapping it to a numerical range of 0-1; The fatigue coefficient is calculated based on the total number of unanswered questions after normalization, the time since the last rest, and the percentage of time spent answering those questions. The formula is as follows: In the formula, The fatigue coefficient, , and These represent the total number of unanswered questions after normalization, the time since the last break, and the percentage of time spent answering. They are respectively , and The preset proportional coefficient, and All are greater than 0.

5. A neonatal intensive care unit service platform according to claim 1, characterized in that, The steps for calculating the empirical coefficient of the response are as follows: The number of times each medical staff member has completely answered family members' questions about newborns from the manual response end is recorded, and the time interval from the start of answering the question to the current time is recorded as the time interval for answering. The number of times a family member provides a complete answer to their questions about the newborn is divided by the time interval during which the answer is provided, and this number is recorded as the answer density value. Obtain the response satisfaction score of each medical staff member after each response is completed, and calculate the average response satisfaction score as the response satisfaction score; The response density and response satisfaction values ​​are normalized, and the response experience coefficient is calculated based on the normalized response density and response satisfaction values. The calculation formula is as follows: In the formula, To answer the empirical coefficient, and These represent the normalized response density value and response satisfaction value, respectively. This is a preset weighting factor, with a value range of 0-1.

6. A neonatal intensive care unit service platform according to claim 1, characterized in that, The steps for calculating the response priority value based on the response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient are as follows: The response priority value is calculated by normalizing the response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient. The steps for calculating the response priority value based on the normalized response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient are as follows: In the formula, In order to respond to the priority value, These are the normalized response timeliness coefficient, response time stability coefficient, fatigue coefficient, and response experience coefficient, respectively. They are respectively The preset proportional coefficient, and All are greater than 0.

Images